Privacy Policy

Version 1.0 · Effective October 25, 2026 · Last updated: October 25, 2026

Summary

Dagenti is a platform for artificial intelligence assistants, such as Lía, that chat with you on WhatsApp.
The assistant notes what you tell it so it can remember it later, without asking you each time. You can see, correct, and delete what it notes.
What the assistant or our filters recognize as sensitive (for example, about your health or money) is stored encrypted and marked. It isn't added to answers on its own: the assistant looks it up when what you ask requires it.
Whoever administers the assistant can read your conversations and what the assistant notes. For Lía, that is us.
We use providers, most of them in the United States: OpenAI and Anthropic (AI models), Meta (WhatsApp), Railway (servers), and the others listed in section 6.
We don't sell your data or use it to train AI models.
Conversations are kept for 12 months on our servers; the copy OpenAI keeps follows its own rules (section 8). What the assistant notes is kept until you or the administrator delete it.
To get a copy of your data, correct it, or delete it, write to contact@dagenti.com.

1. Who we are

Dagenti is operated by Dagenti AI SpA, RUT 78.517.976-5, located at Avenida Providencia 1208, oficina 1607, Providencia, Región Metropolitana, Chile ("we" or "us"). Legal representative: Daniel Millar. Privacy contact: contact@dagenti.com.

In this policy:

Assistant means Lía or any other AI assistant that runs on Dagenti.
Administrator means whoever invites you to use an assistant and configures it. For Lía, that is us. If a company offers you a Dagenti assistant, the administrator is that company (section 13).

2. What data we process

Your number and contact details. The WhatsApp number you write from, the name you give the assistant and, if you choose, your email. Your email is used only after you confirm it through a link. We don't read your WhatsApp profile name.
Your messages. Text, options you tap, locations you share, and voice notes. Voice notes are transcribed and the audio is then discarded: we keep only the text. Today the assistant does not accept videos or documents: it declines them and doesn't store them.
If you send a photo, the assistant looks at it to answer you, but we don't keep the image: before sending it to the AI model, we remove its location and other hidden file data. The conversation records that you sent a photo, what you wrote, and a description of what it shows, written by the AI model. We don't identify people by their faces.
What the assistant notes about you: facts, likes, to-dos, and people you mention (section 4).
Reminders and tasks you ask for.
Apps you connect, such as Gmail or Google Calendar. The assistant can only read from them, and only from apps the administrator allows. What it reads to answer you becomes part of the conversation.
Your browser. If the assistant visits websites and acts on them for you (for example, signing in, or filling out and submitting forms), it uses a browser of your own, which keeps the sessions and cookies of those sites. The passwords you save for those sites (your "keychain") are entered in your profile, never in the chat, and are stored encrypted.
Alerts from other services, such as TradingView or GitHub alerts you set up to receive on WhatsApp, if that feature is available to you.
Your Lía address and the emails that reach it. If you choose a Lía address (for example, name@lia.dagenti.com), anyone can write to you there. For each email we keep, encrypted, for 30 days, the sender, the subject, and a summary written by the AI model; we don't keep its text or its attachments. The conversation records only who it is from and its subject. If you ask the assistant to open an email, it requests the text again from our email provider, within those 30 days, and what it reads to answer you becomes part of the conversation. Invitations that arrive are kept in the assistant's calendar, which confirms only the ones you approve, and the rules you give it ("always add Ana's") are kept in your profile. The assistant doesn't do what someone else's email asks and doesn't tell the sender anything about you.
MCP servers you connect. If you connect an MCP server in your profile (a service that gives the assistant tools), the assistant sends it what it needs for your request, which may include parts of your conversation or of what it has noted about you, and uses what the server returns. You choose that server; whoever offers it runs it, not us, and what it receives is subject to its own terms. We store its address and the key you paste encrypted; what the assistant reads from it stays in the conversation. Before the assistant uses a new or changed tool that only reads, we review it and try it once on your behalf, with made-up data; if the review holds it back, the assistant doesn't use it unless you approve it in your profile. Before using a tool that writes, sends or deletes something, the assistant asks you. You can remove a server in your profile, and "Borrar todo" ("Delete everything") removes them all.
Usage and cost. How many messages you send and how much your use of the AI model costs each month, to apply limits.
If you arrive from an ad on Facebook or Instagram that opens WhatsApp, Meta tells us which ad you came from.
If you write without an invitation, we only keep, for one hour, the fact that we replied to you. During that hour, the administrator can see your number to fix an invitation.

Where it comes from: you; the apps and sites you connect or ask the assistant to visit; the MCP servers you connect; people who write to your Lía address; WhatsApp (Meta), which gives us your number and messages; and the administrator, who invites you using your number.

If you write to a Lía address

Addresses ending in @lia.dagenti.com belong to people who use Lía. If you write to one:

Our email provider, Resend, receives your email in São Paulo (Brazil) and stores its account data in the United States. It keeps what arrives, including what we discard, for 30 days.
We keep, encrypted, for 30 days, your email address, the subject, and a summary of your email written by an OpenAI AI model. We don't keep its text or its attachments. What we discard, such as bulk email, we don't keep: we only count it.
The person receives that summary on WhatsApp or, when WhatsApp doesn't allow us to message them, at their email address. Their conversation with the assistant records your address and the subject, for the conversation retention period (section 9), and whoever administers the assistant can see them (section 5). If the person asks to open your email within those 30 days, we request its text again from Resend, and what the assistant reads becomes part of that conversation.
If your email contains an invitation, the assistant keeps it in its calendar, encrypted, until 30 days after the event, and confirms it only if the person approves.
The assistant doesn't do what your email asks, doesn't reply to you on its own, and doesn't tell you anything about the person.
If you write to postmaster@lia.dagenti.com or abuse@lia.dagenti.com, your address and the subject reach the administrator's team by email.
To ask us to delete what we keep from your email, write to contact@dagenti.com.

3. How we use your data

To chat with you and do what you ask the assistant to do.
To remember what you tell it and your preferences.
To send you reminders and tasks on WhatsApp or, if you gave us your email, by email when WhatsApp doesn't allow us to message you.
To read your connected apps, and to visit sites and act on them, when you ask.
To receive the emails that reach your Lía address, tell you what they bring, and confirm in the assistant's calendar the invitations you approve.
To use the MCP servers you connect when you ask, and to review their new or changed tools before using them.
To alert the administrator's team when the assistant can't resolve something. That email says what you asked for, without your full number.
To review some conversations, with the help of AI, to improve the assistant's instructions.
To protect your account, for example with sign-in alerts.
To manage usage and cost, such as the monthly limit set by the administrator.
To keep the service secure and prevent abuse.

We use your data because it is necessary to provide the service you accept by using it, because we have an interest in keeping it secure, improving it, and managing its cost, and, for optional features (your email, your apps, your browser, your Lía address, your MCP servers), because you turn them on.

We don't use your data for advertising, we don't sell it, and we don't use it to train AI models.

4. What the assistant notes

The assistant notes what you tell it that will help it help you later, without asking you each time.
In your profile you can see, search, correct, and delete everything it has noted. In the chat you can say "what do you know about me regarding…?", "correct…", or "forget that".
Sensitive information. If you tell it something sensitive about yourself, such as health information, salary or debts, beliefs or religion, sex life or sexual orientation, ethnic origin, political views, union membership, or biometric data, it also notes it without asking. Whatever the assistant or our filters recognize as sensitive:
- is stored encrypted and marked as sensitive;
- isn't added to answers on its own: the assistant looks it up when what you ask requires it;
- is not used to build search indexes.
Other people's information. If you share sensitive information about someone else, for example a patient or a colleague, and the assistant recognizes it as someone else's, it is saved only if you expressly ask the assistant to save it. Share it only if you have the right to do so.
Passwords and cards. Don't type passwords, card numbers, or codes in the chat: the assistant doesn't keep them in its notes. Website passwords are entered only in your profile.
What the assistant notes is sent to the AI model when needed to answer you.

5. Who can see your data

You, in the chat and in your profile.
The assistant's administrator (for Lía, us). The administrator can see:
- the full text of your conversations. A person on their team can take over the conversation and reply to you instead of the assistant, in the same chat and without a mark that sets it apart;
- everything noted about you, including sensitive information;
- your reminders, the requests the assistant sent to their team, your name, the last 4 digits of your number, your partially hidden email, and the cost of your use this month;
- which sites your browser visited, but not what you typed on them or your passwords.
- which MCP servers you connected (only their domain) and how many times they were used, but not their full address or your keys.
The administrator can also pause your access, delete your space and, separately, delete your conversations (section 10).
Our technical staff, only when needed to operate, support, or protect the service.
The providers in section 6, only to provide the service.
The MCP servers you connect, with what the assistant sends them for your request (section 2).
Authorities, when the law requires us.

6. Providers

Provider	Purpose	What it receives	Where
OpenAI	Main AI model; transcribing voice notes; reviewing conversations to improve the assistant, when we use its model for that. Also through a ChatGPT subscription plan (section 8)	Your messages, notes needed for the answer, what the assistant reads from your apps, from your MCP servers or from websites, alerts from other services you ask it to review, the audio of your voice notes, the photos you send, and the emails that reach your Lía address, to summarize them and when you ask to open them	USA
Anthropic	Backup AI model: conversations that start while OpenAI is failing are handled entirely by it; reviewing conversations to improve the assistant, when we use its model for that; with the subscription plan (section 8), it also handles any part of a conversation that neither the plan nor OpenAI's API manages to answer	The same as OpenAI, for those conversations, and the conversations reviewed with its model; also the photos you send while OpenAI is failing	USA
Meta (WhatsApp)	Carrying messages and voice notes	Your number and your messages	Meta data centers in the USA, Ireland, Denmark, Sweden, and Singapore
Railway	Hosting our servers, database, and browsers	Everything we store	USA (Virginia)
Composio	Connecting your apps and storing the authorizations you give	A random identifier (never your number) and what the assistant reads from your apps	USA
Resend	Sending emails to you and to the administrator's team; receiving the emails that reach your Lía address	Your email, the notices we send you, and requests to the team (what you asked for, what the assistant checked, and what it proposes, without your full number); the emails that reach your Lía address, including those we discard	USA; emails to your Lía address are received in São Paulo (Brazil), and Resend stores its account data in the USA
The MCP servers you connect	Giving the assistant the tools you choose	What the assistant sends them for your request, which may include parts of your conversation or of what it has noted about you; the key you paste for each one; and, when we review a new or changed tool that only reads, a test call with made-up data	Wherever whoever offers them runs them

Each provider uses your data only to provide its service to us, under its contract with us. We don't sell or give your data to anyone. The MCP servers you connect are not our providers: you choose them, whoever offers them runs them, and what they receive is subject to their own terms.

7. Transfer to the United States

Our servers and most of our providers are in the United States. If you live in another country, your data is sent to and stored there. These transfers are covered by the data processing agreements we have with each provider. What goes through the ChatGPT subscription plan (section 8) is governed by that plan's terms. Emails to your Lía address are received in Brazil (section 6). The MCP servers you connect may be in any country.

8. AI model training

We don't use your conversations or what the assistant notes to train AI models. If we ever wanted to, we would ask your permission first, and would only do it if you agree.
We use AI models through each provider's business services (API). In addition, the assistant first uses an OpenAI subscription plan (ChatGPT) that we hold to generate its answers, search the web, look at the photos you send, and review the alerts from other services you ask it to review; if the plan doesn't answer in time, it continues through the API and, if the API doesn't answer either, it may continue through Anthropic's backup model (section 6). When we choose to, we also review conversations to improve the assistant through that plan. Through that route OpenAI receives the same as through its API, it is governed by that plan's terms rather than its API terms, and in that plan we have turned off the option that lets it use that content to train its models. It also summarizes the emails that reach your Lía address through that route.
OpenAI says it doesn't use data received through its API to train its models unless the customer opts in (we have not), and that it keeps that data for up to 30 days to detect abuse. It also says it doesn't keep audio sent for transcription. OpenAI also keeps a copy of each conversation to keep track of the thread and, according to OpenAI, keeps it until it is deleted. Today we don't delete it: neither deleting your account nor the deletions in section 10 reach it.
Anthropic says that, by default, it doesn't use API data for training and deletes it within 30 days, unless it detects use that violates its policies (in which case up to 2 years) or the law requires it to keep it.
Meta says it keeps business WhatsApp messages for up to 30 days and doesn't automatically use them to choose the ads you see.
These are each provider's own policies and may change. This summarizes what they said on October 25, 2026.

9. How long we keep your data

Data	Period
Conversations	12 months on our servers, including the copy of the dialogue when the backup model or the subscription plan answered. "Borrar todo" ("Delete everything") doesn't delete them; you can also ask us to delete them (section 10). The administrator can also delete them
Copy of the conversation at OpenAI	We don't delete it; according to OpenAI, it is kept until it is deleted (section 8)
What the assistant notes and your preferences	Until you or the administrator delete them
Requests to the team	In your space, until it is deleted; in the administrator's team email, under their policy, with no period set by us
Conversation reviews	Until they drop out of the assistant's 60 most recent reviews. They are deleted with the reviewed conversation and with "Borrar todo"
Your number on the invitation list	Until the administrator removes it, even if you delete your data
If the administrator removes your access	Your space and what was noted are kept until the administrator deletes them. You can ask them to do so
Voice notes	The audio is discarded once transcribed; the text stays with the conversation
Photos	The image is not kept. What you wrote and the AI model's description stay with the conversation
Reminders and tasks	Until you or the administrator delete them
Email	Until you delete it. Every email includes a link to stop receiving them
Browser and keychain	Until you delete them
Sites your browser visited	The last 40 sites, until the browser sessions close, you use "Borrar todo", or your space is deleted
Alerts from other services	7 days
Emails that reach your Lía address	30 days: the sender, the subject, and the summary, encrypted. We don't keep their text or attachments. The conversation records who each is from and its subject, for the conversation retention period
Emails received at Resend	30 days, including those we discard
The assistant's calendar	Until 30 days after each event
Your email rules	Until you delete them
Your Lía address	Until you remove it or your space is deleted. If you change it, the previous one keeps reaching you for 90 days, or until the end of the last event the assistant organized with you if that is later. A removed address is never given to anyone else: we keep only an encrypted fingerprint of it, forever, with nothing that links it to you
MCP servers	Their address and your key, encrypted, and your last 50 uses, until you remove them or choose "Borrar todo". What the assistant read from them stays with the conversation
Result of each tool's review	30 days, with no data of yours
Monthly usage and cost	While your space exists ("Borrar todo" keeps it, to apply the cap)
Ad you came from	The same as your conversations
Link to your profile	Expires in 15 minutes
Numbers that write without an invitation	1 hour
Data held by providers	Under their policies (section 8)

Our server keeps backups: daily ones, kept for 6 days, and weekly ones, kept for 27 days. What you delete leaves them when they expire, at most 27 days later.

10. How to delete your data

In the chat: say "forget that" and the assistant deletes that note or preference.
In your profile: delete items one by one or choose "Borrar todo" ("Delete everything"). It deletes your name, what was noted, your preferences, reminders, your email, your connected apps (we also ask the connections provider to delete them), your browser, your keychain, and your alerts. It doesn't delete your conversation history, which follows the period in section 9, and it can't be undone. It also deletes the reviews of your conversations. It also deletes the emails that reached your Lía address, the assistant's calendar, and your email rules; your address is kept. It also removes your MCP servers.
If you ask the assistant to forget everything, it sends you a link to your profile, valid for 15 minutes, to do it there.
Your Lía address: in your profile, "Quitar mi dirección" ("Remove my address") removes it for good.
MCP servers: in your profile, "Quitar" ("Remove") deletes a server, its key, and its activity.
If you only use WhatsApp and also want your conversations deleted, write to contact@dagenti.com with your number. We do it within 30 days.
The administrator can delete your conversations ("Borrar conversaciones", "Delete conversations"). If they delete your space, what is in it is deleted, but not your conversations, which follow the period in section 9.
Emails: to stop receiving them, use the link included in each one.
What remains after deleting: your number on the invitation list, until the administrator removes it; emails already sent to the administrator's team; the copy of your conversations that OpenAI keeps (section 8); and whatever the other providers already received, which follows their own retention periods.

What each action deletes:

Action	Conversations	Conversation reviews	What the assistant notes and your preferences
"Borrar todo" ("Delete everything"), in your profile	No	Yes	Yes
"Borrar conversaciones" ("Delete conversations"), by the administrator	Yes	Yes	No
The administrator deletes your space	No	Yes	Yes
Automatically, after 12 months (section 9)	Yes	Yes	No

11. Your rights

You can:

know what data we have about you and get a copy;
correct it;
delete it;
object to a use or ask us to restrict it;
receive your data in a format you can take to another service;
withdraw your permission for optional features (your email, your apps, your browser, your Lía address, your MCP servers) by disconnecting or deleting them in your profile.

How: from your profile, in the chat, or by writing to contact@dagenti.com. To confirm it's you, we may ask you to write from your WhatsApp number or use the link to your profile. You can act through an authorized agent.

We respond within 30 days; if we need more time, we'll let you know beforehand, up to 30 additional days. It's free, and we won't treat you differently for exercising your rights. If we deny a request, we'll tell you why. You can ask us to reconsider and, if you still disagree, complain to the data protection authority where you live (section 16).

If the administrator is a company, you can also contact them; if you write to us, we'll pass your request on to them.

12. Security

The passwords in your keychain and the keys for your alerts are stored encrypted. The authorizations for your apps are kept by the connections provider, not by us.
Sensitive information the assistant notes is stored encrypted.
What we keep from the emails that reach your Lía address, and the assistant's calendar, are stored encrypted.
The address of your MCP servers and your keys are stored encrypted. The AI model never sees your keys or the full address: only the name you give each server and its domain.
Conversations and notes that are not sensitive are stored on servers with restricted access.
Access links and codes are stored in a form that can't be read back, and expire quickly.
Website passwords never go through the chat and are never shown.

No system is infallible. If an incident affects your data, we will notify you where required and take steps to fix it.

13. Assistants offered by a company

If a company, such as your employer, offers you a Dagenti assistant, that company is the administrator: it decides what it uses the assistant for and is responsible for your data. We process the data on its behalf, under our contract with it. The company can see what section 5 describes. Please also review that company's own privacy policy.

14. Children

Dagenti is for people 18 and older and is not directed to children. We don't ask your age. If we learn that someone under 18 is using an assistant, we will close their access and delete their data. If you believe a minor has given us data, write to contact@dagenti.com.

15. Website, cookies, and Do Not Track

dagenti.com doesn't use advertising cookies or tools that track you across sites. Your profile and the app use only what they need to work, such as your session. Because we don't do that tracking or sell data, we don't change anything when we receive a "Do Not Track" or "Global Privacy Control" signal.

16. Additional information by location

Chile

Controller: Dagenti AI SpA, RUT 78.517.976-5, Avenida Providencia 1208, oficina 1607, Providencia, Región Metropolitana, Chile, contact@dagenti.com.
Legal bases: providing the service you accept; our legitimate interest in keeping it secure, improving it, and managing its cost; your consent for optional features (email, apps, browser, Lía address, MCP servers). For sensitive data, your consent: you tell them to the assistant on your own initiative, and you can withdraw it by deleting them.
Rights: access, rectification, erasure, objection, temporary blocking, and portability. We respond within 30 calendar days, extendable once by another 30 if we notify you. If we deny your request or don't respond, you can file a complaint with the Agencia de Protección de Datos Personales.
Transfers: to the United States, which has no adequacy determination, under each provider's contractual safeguards.
Automated decisions: the assistant generates its answers automatically, but it doesn't make decisions about you that have legal effects or similarly significant effects on you.
Profiling: the assistant builds a profile from what it notes (likes, preferences, to-dos) only to personalize its answers; it isn't used to make decisions about you or for advertising.

United States

This section applies to the extent your state's privacy laws, such as California's, apply to us.

Categories of data collected in the last 12 months: identifiers (number, name, email, app identifiers, shortened IP address); content of messages, photos and voice notes, and of the emails that reach your Lía address; geolocation (approximate country); internet activity (sites your browser visits for you and account events); preferences and what the assistant notes; and sensitive personal information (sensitive things you share, precise location you share, the contents of emails and messages the assistant reads from your connected apps, and the credentials in your keychain). Sources and purposes are in sections 2 and 3, and retention periods in section 9.
Sale and advertising: we don't sell your data or share it for advertising based on your activity on other sites. We have no knowledge of selling data of anyone under 16.
Sensitive information: we use it only to provide the service you ask for, not to infer characteristics about you.
Service providers: those in section 6, except the MCP servers you connect yourself, for the purposes stated there.
Your rights: to know and access the data we have, delete it, correct it, receive it in a portable format, not be treated differently for exercising these rights, and appeal if we deny your request. You can exercise them as described in section 11, including through an authorized agent.
Automated decisions: we don't use the assistant to make significant decisions about you, such as decisions about credit, employment, housing, education, or health care.
AI assistant: Lía and the other assistants are artificial intelligence, not people. If you ask, they will confirm it. A person on the administrator's team may reply to you instead, in the same chat.

17. Changes to this policy

We will post changes on this page with the new date and version. If they are significant, we'll notify you on WhatsApp or by email 15 days before they take effect. If a change expands how your data is used, for example to train models, we'll ask your permission first.

18. Contact

contact@dagenti.com · Avenida Providencia 1208, oficina 1607, Providencia, Región Metropolitana, Chile.